The following Chinese draft standard proposes rules for generative AI data annotation and data labeling with an eye toward improving the safety and security of GenAI systems. The standard defines safety and security as including not only protection of people’s physical safety and disinformation prevention, but also censorship of content that criticizes Communist Party rule or presents China in an unflattering light. China issued a finalized version of these standards in April 2025, but, as of the publication date of this translation, CSET has not observed a publicly available full-text copy of the final version.

An archived version of the Chinese source text is available online at:    https://perma.cc/QK8D-ZPRB

National Standard of the People’s Republic of China: Cybersecurity Technology – Generative Artificial Intelligence Data Annotation Safety Specifications (Draft for Feedback)

1. Scope

This document specifies the basic safety¹ requirements for data annotation used in the training of generative artificial intelligence (GenAI), the safety requirements for data annotation rules, annotation personnel requirements, data annotation verification requirements, and methods for testing annotation safety.

2. Normative Reference

The contents of the following documents, through normative references in this text, constitute indispensable provisions of this document. Among them, for dated references, only the edition corresponding to that date applies to this document. For undated references, the latest edition (including all amendments) applies to this document.

GB/T 42755-2023: Artificial intelligence — Code of practice for data labeling in machine learning

3. Terminology and Definitions

The terms and definitions listed below apply to this document.

3.1 Prompt

Input information that is used to guide a GenAI model in completing a specific task and generating an appropriate output.

3.2 Response

In GenAI data annotation, a human-understandable reply generated in accordance with the requirements of the prompt. This is used to train the model to output corresponding content, patterns, or styles in response to prompts.

To view the rest of this translation, download the pdf below.

1. Translator’s note: The Chinese word 安全 encompasses the meanings of both “safety” (protection from accidental harm) and “security” (protection from deliberate harm). In this translation, it is variously translated as “safety,” “security,” “safety and security,” or “safety or security” at the translator’s discretion.